Re: [PATCH] x86/sgx: clarify 'laundry_list' locking

[Jarkko Sakkinen]

On Mon, Nov 16, 2020 at 02:25:31PM -0800, Dave Hansen wrote:
> From: Dave Hansen <dave.hansen@xxxxxxxxxxxxxxx>
> 
> Short Version:
> 
> The SGX section->laundry_list structure is effectively thread-local,
> but declared next to some shared structures.  Its semantics are clear
> as mud.  Fix that.  No functional changes.  Compile tested only.
> 
> Long Version:
> 
> The SGX hardware keeps per-page metadata.  This can provide things like
> permissions, integrity and replay protection.  It also prevents things
> like having an enclave page mapped multiple times or shared between
> enclaves.
> 
> But, that presents a problem for kexec()'d kernels (or any other kernel
> that does not run immediately after a hardware reset).  This is because
> the last kernel may have been rude and forgotten to reset pages, which
> would trigger the the "shared page" sanity check.
> 
> To fix this, the SGX code "launders" the pages by running the EREMOVE
> instruction on all pages at boot.  This is slow and can take a long
> time, so it is performed off in the SGX-specific ksgxd instead of being
> synchronous at boot.  The init code hands the list of pages to launder
> in a per-SGX-section list: ->laundry_list.  The only code to touch this
> list is the init code and ksgxd.  This means that no locking is
> necessary for ->laundry_list.
> 
> However, a lock is required for section->page_list, which is accessed
> while creating enclaves and by ksgxd.  This lock (section->lock is
> acquired by ksgxd while also processing ->laundry_list.  It is easy
> to confuse the purpose of the locking as being for ->laundry_list
> and ->page_list.
> 
> Rename ->laundry_list to ->init_laundry_list to make it clear that
> this is not normally used at runtime.  Also add some comments
> clarifying the locking, and reorganize 'sgx_epc_section' to put 'lock'
> near the things it protects.
> 
> Note: init_laundry_list is 128 bytes of wasted space at runtime.  It
> could theoretically be dynamically allocated and then freed after the
> laundering process.  But, I suspect it would take nearly 128 bytes
> of extra instructions to do that.
> 
> Cc: Jethro Beekman <jethro@xxxxxxxxxxxx>
> Cc: Serge Ayoun <serge.ayoun@xxxxxxxxx>
> Cc: Jarkko Sakkinen <jarkko@xxxxxxxxxx>
> Signed-off-by: Jarkko Sakkinen <jarkko@xxxxxxxxxx>
> Signed-off-by: Dave Hansen <dave.hansen@xxxxxxxxxxxxxxx>

Thansk, this does make sense to me. I'll squash it.

/Jarkko

> ---
> 
>  b/arch/x86/kernel/cpu/sgx/main.c |   14 ++++++++------
>  b/arch/x86/kernel/cpu/sgx/sgx.h  |   15 ++++++++++++---
>  2 files changed, 20 insertions(+), 9 deletions(-)
> 
> diff -puN arch/x86/kernel/cpu/sgx/main.c~sgx-laundry-comments arch/x86/kernel/cpu/sgx/main.c
> --- a/arch/x86/kernel/cpu/sgx/main.c~sgx-laundry-comments	2020-11-16 13:55:42.624972349 -0800
> +++ b/arch/x86/kernel/cpu/sgx/main.c	2020-11-16 13:58:10.652971980 -0800
> @@ -36,13 +36,15 @@ static void sgx_sanitize_section(struct
>  	LIST_HEAD(dirty);
>  	int ret;
>  
> -	while (!list_empty(&section->laundry_list)) {
> +	/* init_laundry_list is thread-local, no need for a lock: */
> +	while (!list_empty(&section->init_laundry_list)) {
>  		if (kthread_should_stop())
>  			return;
>  
> +		/* needed for access to ->page_list: */
>  		spin_lock(&section->lock);
>  
> -		page = list_first_entry(&section->laundry_list,
> +		page = list_first_entry(&section->init_laundry_list,
>  					struct sgx_epc_page, list);
>  
>  		ret = __eremove(sgx_get_epc_virt_addr(page));
> @@ -56,7 +58,7 @@ static void sgx_sanitize_section(struct
>  		cond_resched();
>  	}
>  
> -	list_splice(&dirty, &section->laundry_list);
> +	list_splice(&dirty, &section->init_laundry_list);
>  }
>  
>  static bool sgx_reclaimer_age(struct sgx_epc_page *epc_page)
> @@ -418,7 +420,7 @@ static int ksgxswapd(void *p)
>  		sgx_sanitize_section(&sgx_epc_sections[i]);
>  
>  		/* Should never happen. */
> -		if (!list_empty(&sgx_epc_sections[i].laundry_list))
> +		if (!list_empty(&sgx_epc_sections[i].init_laundry_list))
>  			WARN(1, "EPC section %d has unsanitized pages.\n", i);
>  	}
>  
> @@ -632,13 +634,13 @@ static bool __init sgx_setup_epc_section
>  	section->phys_addr = phys_addr;
>  	spin_lock_init(&section->lock);
>  	INIT_LIST_HEAD(&section->page_list);
> -	INIT_LIST_HEAD(&section->laundry_list);
> +	INIT_LIST_HEAD(&section->init_laundry_list);
>  
>  	for (i = 0; i < nr_pages; i++) {
>  		section->pages[i].section = index;
>  		section->pages[i].flags = 0;
>  		section->pages[i].owner = NULL;
> -		list_add_tail(&section->pages[i].list, &section->laundry_list);
> +		list_add_tail(&section->pages[i].list, &section->init_laundry_list);
>  	}
>  
>  	section->free_cnt = nr_pages;
> diff -puN arch/x86/kernel/cpu/sgx/sgx.h~sgx-laundry-comments arch/x86/kernel/cpu/sgx/sgx.h
> --- a/arch/x86/kernel/cpu/sgx/sgx.h~sgx-laundry-comments	2020-11-16 13:55:42.627972349 -0800
> +++ b/arch/x86/kernel/cpu/sgx/sgx.h	2020-11-16 13:55:42.631972349 -0800
> @@ -34,15 +34,24 @@ struct sgx_epc_page {
>   * physical memory e.g. for memory areas of the each node. This structure is
>   * used to store EPC pages for one EPC section and virtual memory area where
>   * the pages have been mapped.
> + *
> + * 'lock' must be held before accessing 'page_list' or 'free_cnt'.
>   */
>  struct sgx_epc_section {
>  	unsigned long phys_addr;
>  	void *virt_addr;
> -	struct list_head page_list;
> -	struct list_head laundry_list;
>  	struct sgx_epc_page *pages;
> -	unsigned long free_cnt;
> +
>  	spinlock_t lock;
> +	struct list_head page_list;
> +	unsigned long free_cnt;
> +
> +	/*
> +	 * Pages which need EREMOVE run on them before they can be
> +	 * used.  Only safe to be accessed in ksgxd and init code.
> +	 * Not protected by locks.
> +	 */
> +	struct list_head init_laundry_list;
>  };
>  
>  extern struct sgx_epc_section sgx_epc_sections[SGX_MAX_EPC_SECTIONS];
> _
>